Reproduce “Spatio-Temporal Attention for Consistent Video Semantic Segmentation in Automated Driving” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Serin Varghese, Kevin Ross, Fabian Hueger, Kira Maag
STA improves temporal consistency in video semantic segmentation by letting current-frame attention use features from recent frames. It combines keys and values across a short history, without explicit optical-flow alignment in the model. On BDD100k, STA-SegFormer B0 raises mean temporal consistency from 71.40% to 80.60%, while mean intersection over union changes from 51.40% to 51.64%, versus its single-frame baseline.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.